Icelake with integrated TB3 controllers completely crushes non-Icelake CPU's with external controllers in performance (up to 40% gains...)
I was alerted to Jarod Tech's findings on NotebookCheck and immediately jumped here for the discussion. I'm very keen to setup a solid ultrabook eGPU setup for when I'm working on-site with an RV. This is encouraging!
I'm also very curious how Thunderbolt 4 will play out. Looking forward to seeing those benchmarks.
I got an HP Spectre X360 [15-eb1043dx] yesterday with the i7-1165G7 processor. Thanks to the 15-in footprint and no dGPU, the cooling system is working surprisingly well. The CPU is set at 28W and can run consistently at 3.3GHz and temps hovering around low 70s Celsius. I let AIDA64 System Stability test run for over an hour and the laptop was not even warm to the touch. As seen in the photo below, the warmest spot in the middle top half of the keyboard was mid 30s Celsius. Noise was not bad either at low 50s dB.
Thunderbolt 4 is not necessarily faster than Thunderbolt 3. As with most things we've learned about this technology, every component of an eGPU setup can make a difference. You'd want the best host laptop possible in order to extract the most performance out of the external graphics card. The go-to configuration is direct Thunderbolt connection to the CPU. Ice Lake and now Tiger Lake provide that ideal connection. With the exception of Alienware and Apple 15-in/16-in MacBook Pro, most H-CPU laptops route their Thunderbolt connection through the PCH and the manufacturers don't care how it performs in eGPU mode.
Next is the CPU TDP and its cooling system. I'm a firm believer in iGPU-only ultrabook. There's only one main component to cool and it keeps things simple regarding graphics switching. This HP Spectre X360 15 is a very good example of what a perfect eGPU host in late 2020 should be. I will post a build guide soon with benchmark numbers.